Identifying illicit transactions in Bitcoin Tumbler services using Supervised Machine Learning Algorithms

Abstract

Digital currencies have emerged as a new platform for money laundering as the first blockchain platform to support smart contracts, and the number of transaction records for Bitcoin has expanded substantially in recent years. Due to privacy services such as tumblers, which conceal the identities of payers and payees, pseudonymous digital currencies are frequently referred to as anonymous address and untraceable. A large proportion of accounts participating in Bitcoin transactions are engaging in illegal activity. According to the digital ledger, a transaction was transferred from one of many potential payers to one of many possible payees. As a consequence, it is vital to have a dependable system for classifying accounts and keeping track of the transactions linked with each account. We utilized machine learning techniques to apply multi-class classification methods to a Bitcoin transaction record obtained from kaggle.com in order to improve the model’s accuracy. Experiment results show that ML algorithms like logistic regression, random forests, and multilayer perceptron’s with the macro-average F1 score for bitcoin tumblers outperform state-of-the-art approaches for identifying unknown digital currency accounts, with the RF model outperforming LR and MLP by 97% accuracy.